DeepSeek V4-Pro announces permanent price reduction: API output cost is less than 30 TWD per million tokens

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1578 words

On May 22, 2026, DeepSeek announced that the API price for its V4-Pro model would be permanently reduced to 6 RMB. As AI Agent software rolls out on a global scale, the demand for computing power has exploded, leading most AI vendors to raise prices. DeepSeek, however, has chosen to go against the grain. (Previous coverage: DeepSeek V4 released: Can its coding capabilities beat GPT and Claude? Costs once again top the charts) (Background: Why does the extreme cost-effectiveness of Chinese AI make Silicon Valley collectively anxious?) DeepSeek announced this week that it is making its limited-time discount permanent. After the promotion ends on May 31, the output pricing per million tokens will remain at 6 RMB (approximately 28 NTD), effective permanently. Note: The DeepSeek V4-Pro API was previously in a promotional period with a discount of 25% of the original price. According to previous market expectations, the promotional activity was set to end in June 2026, with prices "reverting to the original level." Input costs also remain low. According to this pricing announcement, V4-Pro's input pricing will also be maintained at the promotional level, placing the overall API rate in the lowest tier among mainstream commercial models. For average individual users, the DeepSeek website and App are currently free. The direct beneficiaries of this price reduction are developers and enterprise user groups, especially those with high call volumes who are sensitive to token rates. Why is DeepSeek permanently lowering prices while others are raising them? There are three levels of answers to this question. Level 1: A structural decline in computing costs. The DeepSeek V4 series has been confirmed to run entirely on Huawei Ascend 950 PR chips, moving away from direct reliance on NVIDIA GPUs. Against the backdrop of China's self-built computing ecosystem, DeepSeek's inference cost curve is asymmetric to that of Western AI vendors. While OpenAI and Anthropic must bid for high-priced H100/H200 chips in the market, DeepSeek's hardware procurement path is different, and its cost structure allows for a more aggressive pricing strategy. Level 2: Market share strategy. API price reduction is the most direct means to attract a developer ecosystem. In code capability tests in April, DeepSeek V4-Pro demonstrated the ability to compete head-on with the GPT-4o and Claude Sonnet series. With the gap in model capabilities narrowing, the pricing gap directly determines which platform becomes the developer's default choice. Being 36 times cheaper per million tokens means that for a commercial application with token call volumes in the tens of millions to hundreds of millions, annual API costs drop from millions of dollars to tens of thousands—a watershed moment for procurement decisions. Level 3: A deeper signal. Solidifying a temporary promotion into standard pricing is a gesture of commitment to the market. DeepSeek is telling developers: You can write this pricing into your business plan; it won't suddenly go back up. This is especially important for enterprise clients who require long-term planning. The combination of these three factors presents a clear strategic image: DeepSeek is using pricing as a tool to build developer dependency, with the goal of making Chinese-produced foundation models the underlying inference engine for global commercial AI applications. The standard version of OpenAI GPT-5.5 is priced at $30 per million output tokens, equivalent to approximately 216 RMB. Compared to DeepSeek V4-Pro's 6 RMB, the gap is over 36 times. The GPT-5.5 Pro version is priced at $180, equivalent to approximately 1,296 RMB. Compared to the same 6 RMB, the gap is over 200 times. If a startup team's monthly AI inference cost on OpenAI is $100,000, switching to DeepSeek V4-Pro's theoretical rate could bring it down to under $2,800 (provided, of course, that the model capabilities are interchangeable for their specific application scenarios). Although GPT-5.5 and Claude Opus still have their own moats in complex reasoning, multilingual understanding, and long-text processing, for a large portion of enterprise AI scenarios—such as customer service, document processing, coding assistance, and data summarization—the performance of DeepSeek V4-Pro is already sufficiently powerful. In 2026, a year when computing costs are generally rising, a permanent price reduction is an aggressive political statement: DeepSeek intends to establish its position in the global AI infrastructure layer, and pricing is the fastest ticket to entry.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-21

Get free NVIDIA model API keys! Register in 3 minutes to access Kimi, DeepSeek, and Llama for all your needs.

Similarity 140%關鍵字 api/deepseek同分類 zh

2026-04-21

Good news for Google Pro and Ultra: subscribers get 500 quota in AI Studio, bypassing API fees

Similarity 140%關鍵字 pro/api同分類 zh

2026-05-26

Bloomberg: China "restricts top AI talent from leaving the country," executives at Alibaba and DeepSeek forced to surrender passports

Similarity 120%關鍵字 deepseek同分類 zh

2026-05-26